Perception &: Psychophysics

1992. 51 (5), 465-472

The role of attention

in different visual-search tasks
MARY J. BRAVO and KEN NAKAYAMA
Smith-Kettlewell Eye Research Institute, San Francisco, California

Observers viewed displays containing a variable number of distractors of one color and a tar·
get of another color. In some experiments, the target and distractors maintained their color from
trial to trial; in others, they reversed unpredictably. Observers made a speeded two-choicejudg·
ment concerning either the presence, the color, or the shape of the odd-colored target. With only
one exception, all of these conditions produced the same pattern of results: reaction times reo
mained constant as the number of distractors increased. The exceptional result occurred when
observers judged the shape of the odd-colored target and the color of the target and distractors
reversed unpredictably. In this case, reaction times decreased as the number of distractors in-
creased. These results are interpreted in terms of the attentional requirements of the different
judgments and the mechanisms that guide attention.

Figure 1 shows an odd target located among homoge-
nous distractors. Two observations have been made con-
cerning such displays. First, the odd element' 'pops out, "
that is, it immediately attracts attention. And second, the
target is readily detected regardless of the number of
distractors in the display (Donderi & Zelnicker, 1969;
Egeth, Ionides, & Wall, 1972; Neisser, 1963; Treisman
& Gelade, 1980). It has been assumed that these two ob-
servations are causally related: the target is easy to de-
tect because it immediately summons attention (Duncan
& Humphreys, 1989; Koch & Ullman, 1985; Yantis &
Ionides, 1984).
If we compare the predictions of models of attentional
guidance with the results of detection experiments, this
relationship is less clear. As detailed below, these models
predict that under some circumstances the target should
become easier to find as more distractors are added to
the display. However, the results of detection experiments
do not support this prediction. When the target and dis-
tractors have very different features, as in Figure 1, de-
tection times are found to be unaffected by increasing
numbers of distractors. When the target and distractors
have similar features, detection times are found to in-
crease with increasing numbers of distractors (Duncan &
Humphreys, 1989). Thus, the slope of detection times
plotted against number of distractors is usually either zero
or positive, but rarely negative (for an exception, see
Bacon & Egeth, 1991). Since the models predict a nega-
tive slope, these empirical results indicate that either the
This work was supported by Grant IF32 EY06155 from the NEI and
Grant 83-0320 from the AFOSR. We are indebted to Jeremy Wolfe and
Vera Maljkovic for their comments on an earlier version of this manu-
script. K. Nakayama is now in the Department of Psychology at Harvard
University. Correspondence should be addressed to M. J. Bravo, Smith-
Kettlewell Eye Research Institute, 2232 Webster St., San Francisco,
CA 94115; e-mail: mary@skivs.ski.org.
models are incorrect or that they do not apply to target
detection.
Our hypothesis is that the models of attentional guidance
are correct, but that they do not apply to detection. In other
words, detection does not require attention to be focused
on the target. Of course, some decisions do require selec-
tive attention to be directed to the target, and it is these
tasks that should show the pattern of reaction time data
predicted by the models of attentional guidance. We tested
this conjecture by comparing the predictions of these
models with the pattern of search times for various tasks
performed with identical displays. Our expectation was
that these different tasks would produce different reac-
tion time results and that only tasks requiring focal atten-
tion would produce the results predicted by the models.
Before describing our experiments, we will discuss some
models of how an odd target attracts attention.
ATTENTION-GUIDING MECHANISMS
There are two general ways in which attention may be
directed to an odd target (Duncan & Humphreys, 1989;
LaBerge & Brown, 1989; Wolfe, Cave, & Franzel, 1989).
When observers are looking for a particular target (say,
a red diamond), they can use their knowledge of the tar-
get's characteristics to guide attention; the control of at-
tention has a large top-down component. When observers
are looking for an odd element but do not know its charac-
teristics, then the stimulus must playa greater role in guid-
ing attention. In this case, the control of attention is largely
bottom-up. Although we will use the terms top-down and
bottom-up to label these two types of attention-guiding
mechanisms, we do not mean to imply that one relies
solely on cognitive information and the other solely on
stimulus information. Rather, the labels refer to the relative
weights of these two sources of information. Below we
describe these two types of attention-guiding mechanisms.

465 Copyright 1992 Psychonomic Society, Inc.

466 BRAVO AND NAKAYAMA
Top-Down Mechanisms
Treisman and Gelade (1980) suggested that the visual
system represents the distrib!ltion of each perceptual fea-
ture in a functionaIfy separate map of visual space. There
are maps for vertical, green, movement to the right, etc.
Drawing on this notion, LaBerge and Brown (1989) pro-
posed that if an observer knows the target's unique fea-
ture, then he/she could use the appropriate feature map
to guide attention. For example, if the target is known
to be the only red element in the display, then the locus
of activity among red detectors may guide attention. The
activity in other feature maps is ignored. Thus, if the tar-
get and distractors are so different as to stimulate differ-
ent sets of feature maps, then this strategy will work
equally well regardless of the number of distractors in the
display. However, this strategy requires that the observer
know the target's unique feature. The following bottom-up
mechanisms can operate in the absence of such knowledge.
Bottom-Up Mechanisms
Koch and Ullman (1985) suggested that local inhibi-
tion within feature maps is responsible for the conspicuity
of the target and its subsequent status as the focus of at-
tention. A similar idea has been proposed by Treisman
(1988) in her revised feature-integration model of visual
search. If several distractors share a feature (say, the same
color), the activity they generate in the corresponding fea-
ture map will be suppressed through some form of mutual
inhibition. (For a review of the physiological evidence
for such inhibition see Allman, Miezen, & McGuiness,
1985.) No such suppression will occur to the activity
generated by the target's unique feature because it is not
subject to inhibition generated by other elements sharing
this feature. The activity at each location is then summed
across maps, and attention is directed to the most active
location. Since the inhibitory connections between detec-
tors in each feature map are thought to be local, it is rea-
sonable to suppose that this process is most effective when
the distractors are close together.
An alternative bottom-up mechanism comes from
Julesz's proposal that pop out is the result of processes
that detect feature gradients, or differences in the features
of neighboring elements (Julesz, 1986). This idea posits
mechanisms that compare the features of neighboring ele-
ments and then direct attention to the location of feature
gradients. Because the gradient detectors operate only over
a limited region of space, this process is most effective
when the physical distance between distractor and target
elements is small.
Since both bottom-up mechanisms involve local com-
putations (between target and distractor in one case and
between distractors in the other), they both operate most
effectively when the spacing between elements is small.
Thus, when attention must be directed to an unknown tar-
get, search times should depend on the spacing between
elements in the display. If display size is held constant,
as in these experiments, then the average spacing will
decrease as the number of distractors increases. These
bottom-up models predict that for tasks that require focal
attention to be directed to a target with an unknown unique
feature, search times should decrease as the number of
distractors increases.
We now summarize our predictions for search tasks in-
volving a target with a unique feature. (1) Two patterns
of results are predicted for tasks requiringfocal attention:
(a) When the target's unique feature is known, reaction
times should be independent of the number of distractors.
(b) When this feature is not known, reaction times should
decrease with increasing numbers of distractors. (2) The
models mentioned above make no predictions for tasks
that require distributed attention. We suggest that detec-
tion is one of these tasks, and we know from previously
reported detection experiments that whether the target is
known or unknown, reaction times will be unaffected by
distractor number.
EXPERIMENT 1
Five Search Tasks
This series of experiments (Experiments 1A-I E) exam-
ines the pattern of reaction times for several search tasks.
So that they can be readily compared, the search experi-
ments used the same displays and the same observers. The
experiments differed only in the task the observer per-
formed. The tasks varied in two ways. First, they dif-
fered in the two-<:hoice judgment the observer made about
the odd~olored target; the observer judged either the odd-
colored target's presence, its color, or its shape. These
three judgments are based on information at different spa-
tial resolutions and so are likely to differ in their require-
ments for distributed versus focal attention. Second, we
changed the observer's prior knowledge of the target's
distinguishing feature. We accomplished this by keeping
the target and distractor colors the same from trial to trial
in some experiments and by reversing them unpredictably
in others. These two conditions are termed consistent map-
ping and variable mapping, after Schneider and Shiffrin
(1977). These two methods for sorting the displays might
produce a variation in the strategy the observer used to
search for the target. In the consistent-mapping condition,
the observer could search for a particular feature, whereas
in the variable-mapping condition, the observer must search
for the odd element. As depicted in Table I, the two map-
ping conditions combined with the three judgments pro-
duced a total of five experiments. Note that the cell cor-
responding to consistent mapping/color judgment is blank
because the color judgment is meaningless when the ob-
server knows beforehand the target's color.
Table 1
Five Search Tasks
Mapping
Judgment Variable Consistent
Shape a b
Color c
Detection d e

Method
Observers. Three practiced I psychophysical observers volun-
teered for these experiments. Only one of the observers, K.N., was
aware of the purpose of the experiments.
Stimuli. The displays contained red and green elements drawn
on the black background of a computer monitor. The approximate
luminance of the elements was 10 cd/m' , and that of the background
was 0.7 cd/m' . To ensure that the red and green elements were es-
sentially equiluminant so as not to bias the detection of one or the
other, we used heterochromatic flicker photometry. Before par-
ticipating in the experiments, each observer viewed a I ° square
that alternated in color between red and green. While fixating a
spot 3 ° from the square, the observer adjusted the luminance of
the red to minimize the perceived flicker of the square. The ob-
server's equiluminance setting was then used to color the red and
green elements for that observer.
The colored elements were diamonds (0.8° x 1°) with 0.25° cut
off of either the right or left comer. To ensure that these two shapes
(left cut and right cut) were discriminable when presented at the
eccentricities used in this experiment (4.75°-8.4°), we presented
two observers with the experimental displays and monitored their
eye movements. Both observers were able to confine their eye move-
ments to within I ° of a central fixation spot and at the same time
perform the shape discrimination with high accuracy (>95%).
The colored elements were drawn in randomly selected cells of
a 6 x 8 array (15.2" x 12.6°). The position of the elements within
the cells was jittered by 0.33°. Each display contained 3, 5, 9, 13,
25, or 48 elements, and all but one (the target) had the same color.
The position of the target was chosen randomly with the restric-
tion that it not occur in the two columns closest to fixation. Fig-
ure I shows schematic representations of two such displays. The
detection-task experiments required generating additional displays
in which the target was replaced by a distraetor on 50% of the trials.
In all five experiments, the number of distraetors was varied from
trial to trial. Display size was held constant, so this variation in
the number of distraetors also resulted in a variation in the average
distance between elements in the display.
Procedure. The same procedure was used in the five search exper-
iments, with the only differences among experiments being (I) the
rule for mapping red and green onto the target and distractors and
(2) the two-alternative forced-ehoice judgment the observer was to
make. The procedural similarities are described first.
In all five experiments, the observers viewed the computer-
generated displays from a distance of 70 em. A small fixation cross
was always visible at the center of the display screen, and the ob-
servers were asked to maintain fixation throughout each experiment.
Stimuli were presented with a variable interstimulus interval (lSD,
and the observers responded via a two-button mouse. The display
was presented until the observer responded. When the observer was
correct, the reaction time was recorded. When the observer was
incorrect, feedback was given and an error was recorded. Reac-
tion times on incorrect trials were discarded, and the trials were
•
•• •• •••
<)
•• • ••
••
•• .+. • •
•• •• ••• •••
• ••
+
()
()
• • • •••••••••••
Figure I. Schematics of representative stimuli for two distractor
densities. Actual stimuli consisted of red and green shapes drawn
on a black background.
ROLE OF ATTENTION IN VISUAL SEARCH 467
repeated later in the session. Reaction times less than 200 rnsec or
greater than 2 sec were also discarded, and such trials were repeated
later. Each experiment included an initial practice period involv-
ing typically 240 trials. An experiment consisted of 480 correct trials
(80 per level of distractor number) and was run in blocks of 80
trials, with 4 practice trials preceding each block.
The experiments differed in the following ways. Within an ex-
periment, the colors red and green were assigned to the target and
distractors with either variable or consistent mapping. In the variable-
mapping condition, displays with a red target and green distraetors
were randomly interleaved among displays with a green target and
red distractors. Thus, on any given trial, the observer did not know
beforehand the color of the target or the distraetors. In the consistent-
mapping condition, these two display types were run in separate
blocks so that before each trial the observer knew the color of the
target and distractors. In addition to the different mapping condi-
tions, the experiments differed in the speeded forced-ehoice deci-
sion the observer was asked to make. The decision was always be-
tween two alternatives: either (1) present or absent in the detection
experiment, (2) red or green in the color-identification experiment,
or (3) left cut or right cut in the shape-identification experiment.
The observers pressed the left mouse button to signal "present,"
"red," or "left cut" and the right mouse button to register "ab-
sent," "green," or "right cut."
Results
The results for each of the three observers in the five ex-
periments are shown in Figure 2. Results from consistent-
mapping experiments are plotted with open circles; results
from variable-mapping experiments are plotted with filled
circles. The top two curves show the data from the shape-
identification task, the middle curve shows the data from
the color-identification task, and the bottom two curves
show the data from the target-detection task when the tar-
get was present. We include the data from each subject
to show the remarkable consistency of the results across
observers. The pertinent observation concerns the shape
of these curves. Reaction times decrease as the number
of distractors increases for the shape-identification,
variable-mapping condition (top curve), and reaction times
remain constant for all other tasks. Figure 3 Oeft) shows
the average for the three observers.
Figure 3 (middle) shows the averaged results for the
target-detection task when the target was absent. These
data are plotted in a separate graph because search for
an absent target is logically very different from search
for a present target.
Figure 3 (right) shows the average error rates for the
three observers. To determine whether a speed-accuracy
tradeoff is responsible for the decrease in reaction time
seen in the topmost curve in Figure 3 Oeft), we exam-
ined the correlation between error rates and reaction times
for this condition. No correlation was found, which indi-
cates that the observed reaction-time effect is not due to
observers trading speed for accuracy as the number of
distractors changed.
Discussion
Our main finding is that when observers are asked to
judge the shape of the odd-eolored target and the colors
of the target and distractors reverse unpredictably, reac-
tion times decrease as the number of distractors increases.
468 BRAVO AND NAKAYAMA

700
AR 700 ,
KN
~ shape
g +-
\
t ..
600 i... 600 't----.. ------_.
~
~ \ -6--------.
OO-o,~::::~-------Q
00- (H~---o-- - -- ---0
color
.g 500 +tt... --~------ ... 500 •
...... ---. ---- _--I

~
~
400
.. ,+----6--------.
&>-l.o----o--------\l 400
....... --.. - ------.
00.0-0-
-
0--------0
.
400 • ~ _--.. ------.. detection
..... 0--------<)
oo-u a ---
300 ..........~-~~-~~
3000 10 20 30 40 SO o 1020 3040 SO 3000 10 20 30 40 SO
Number of Distraetors Number of Distractors Number of Distractors

Figure 2. Average reaction time data for each of the three observers in the five different tasks. In each graph, the top
two curves show data from the shape-identification judgment, the middle curve shows data from the color-identification
judgment, and the bottom two curves show data from the target-cietection judgment. Open circles correspond to
COIL'Iistent-mapping experiments and filled circles to variable-mapping experiments. Error bars show one standard error.

This effect is found only when the task involves the
variable-mapping condition and the shape-discrimination
judgment. What is unique about the variable-mapping con-
dition of the shape judgment? We will argue first, that
the spatial resolution required to perform the shape judg-
ment requires focal attention, and second, that the un-
predictable changes in the target's unique feature neces-
sitate a largely bottom-up guidance of attention.
There is considerable empirical evidence suggesting that
selective attention may trade off resolution and spatial ex-
tent. That is, it is possible to attend to a large visual area
with low resolution or a small visual area with fme reso-
lution (Bergen & Julesz, 1983; Eriksen & St. James,
1986). The necessity for this tradeoff is attributed to the
limited amount of information that may be selected by at-
tention for higher level processes like pattern recognition.
We assume that different judgments require different
amounts of spatial resolution. Consider the detection judg-
ment under the variable-mapping condition, where the tar-
get and distractor colors are reversing unpredictably.
Here, the target is defined in relation to the other elements
in the display: the target is odd because it is different from
the other elements in the display, which are all the same.
Simply noting the presence of an odd target could be done
without focusing attention on the target. In fact, focusing
attention on the target gives no additional information for
this task. Similarly, the color-identification judgment could
logically be accomplished without directing attention to
the target. For example, if observers base this decision
on a comparison of the number of occurrences of the dif-
ferent colors, there is no obvious benefit in focusing atten-
tion on the target. However, in the shape-discrimination
experiment, the observer must determine which element
is the target and report about a second feature of that ele-
ment. In this case, there would be a benefit in focusing
attention on the target.

0.2
shape

0.1 ~ /·------~::::::::== ••• o

rP,I'·..••····
700 ~ O.O'--'--'-............~.........--'--...L-J
~. shape 0.2

:::~~~~~~~:~~~~:::::::::::::~
color

~
§.
~... ....-•........~~~~
_
detection
.. detection
(target absent)
0.1.\ ,......

•
-e.-

0.0L................-..-.........
•

-L-..........~...J

'Q 0.2
~ 400 .
~ cP:.()'·······Qo·················O 1P :::i'::::: ..
0.1
• ~. ..~===:=:::::::- ...
300 L...o.-J..~.L-<--..L~-'- ........ '0:1" -..
01020304050 o 10 20 30 40 50 0.0 10 20 30 40 50

Number of Distractors Number of Distractors Number of Distractors

Figure 3. Left: Reaction time data shown in Figure 2, averaged across observers. Open circles correspond to
consistent-mapping experiments and filled circles to variable-mapping experiments. Error bars show one standard
error. Middle: Average reaction time for the detection task when the target was absent from the display. Right:
Average error rates for the three observers.

But the effect cannot simply be ascribed to the involve-
ment of a small attentional aperture in shape discrimination
since the effect did not appear under a different condition
involving this judgment. When the shape-discrimination
judgment was made under the consistent-mapping condi-
tion, reaction times were unaffected by the number of dis-
tractors. Thus, the other critical requirement for the
distractor-number effect is that the observer not know the
target's color before the trial. We believe that the signif-
icance of this requirement relates to the two classes of
attention-guiding mechanisms discussed in the introduc-
tion. Recall that top-down guidance can direct attention
to the target when the target's unique feature is known,
and that the effectiveness of this mechanism is expected
to be independent of the number of distractors. Bottom-
up guidance is required when the target's unique feature
is not known, and this mechanism depends on the num-
ber of distractors.
Thus, we claim the variable-mapping/shape-judgment
task is unique because it requires an attention-guiding
mechanism that is driven by largely bottom-up informa-
tion. However, the shape judgment may have also been
unique in that the relatively high resolution required to
make this judgment may have caused the subjects to make
eye movements. Since we did not monitor fixation dur-
ing these search tasks, it is possible that our observers
made eye movements during the shape-judgment experi-
ment but not during the other experiments. Thus, it could
be argued that the different reaction-time pattern obtained
for these judgments is related to the occurrence of eye
movements for the shape judgments and the absence of
eye movements for the other judgments.
Would this alter our interpretation ? We think not. If we
assume for purposes of argument that the subjects made
saccades to the target location only when asked to judge
the shape of the target, we need to consider if this would
change our explanation of the distractor-number effect.
First, we know that the observed effect of distractor
number cannot simply be due to the putative eye move-
ment itself, since there was no such effect when the sub-
ject made a saccade to a target whose distinguishing fea-
ture was known. The effect must instead be related to
directing a saccade to an unknown target. Previously, we
had assumed that because the shape-discrimination task
requires relatively high resolution, it involves a shift in
focal attention to the target. If we replace this assump-
tion with the assumption that saccades involve a shift in
focal attention to the target, then we are left with our origi-
nal conclusion: tasks that require focal attention give the
pattern of results predicted by the attention-guiding mech-
anisms. Our new assumption, that saccades are preceded
by a shift in focal attention, is commonly made and has
some empirical support (Fischer & Breitmeyer, 1987;
Groner & Groner, 1989; Mayfrank, Mobashery, Kimmig,
& Fischer, 1986; Posner, 1980; Remington, 1980). For
the remainder of this paper, we will assume that the ob-
servers were not making eye movements, but our interpre-
tation of these experiments would not change substantially
if we were to assume the opposite.
ROLE OF ATTENTION IN VISUAL SEARCH 469
EXPERIMENT 2
Replication of Set-Size Effect
With Gabor-Patch Stimuli
The data from the first set of experiments show that
the effect of distractors on visual search may depend on
the observer's task. This suggests that "finding" a fea-
ture target in a detection experiment is different from
"finding" a feature target in order to identify a subtle
feature. We have argued that this difference has to do with
whether the task requires focal attention to be directed
to the target. Our results also suggest that different
attention-guiding mechanisms may be invoked depend-
ing on the observer's prior knowledge of the target's
characteristics. To examine the generality of our results,
we repeated the variable-mapping and consistent-mapping
experiments using different stimuli. Observers were again
asked to make a fine spatial judgment concerning a non-
unique feature of the odd element.
Method
We generated new stimuli consisting of Gabor patches 2 of either
high or low spatial frequency and with a vernier offset to the left
or right. Spatial frequency replaced color as the unique feature,
and the direction of a vernier offset replaced shape as the second,
nonunique feature of the target. Observers were required to deter-
mine whether the vernier offset was to the right or left in the patch
with the odd spatial frequency. Figure 4 shows schematic repre-
sentations of two of these displays. The procedure was the same
as in Experiment I, and the details of the stimuli are given below.
The target and distractor elements were vertically oriented Gabor
patches with spatial frequency centered on 3 cycles per degree or
on I cycle per degree. Each patch had a vernier offset 4 min to
the left or right. These patches were drawn in the cells of a 4 x 3
array (15.2° x 12.6°) with a jitter of 0.66°. Displays contained 2,
3, 4, 6, or 12 elements.
Two observers participated in this experiment, one of whom had
been involved in the previous experiment. 1lte other observer, R.O.,
was paid to participate. R.O. did not know the purpose of the ex-
periment and had never before acted as an experimental observer.
Results and Discussion
The results of the two observers are given in Figure 5.
Although the two observers have very different reaction
times, possibly due to different levels of experience in
psychophysical tests, both show the same pattern of
results. As with the previous experiment, reaction times
decrease as the number of distractors increases only when
the distinguishing features of the target and distractor are
,'~ 'I' 'I' '1'
+
""f ,'~ ,'~
Figure 4. Schematic of two stimuli composed of Gabor funcdons.
In the actuaI experiment, luminance varied more smoothly as It was
the product of a sinusoId and a Gaussian waveform.
470 BRAVO AND NAKAYAMA
700 900
U 650
KN
850
!6OO t.
"''t--+------ i
800
~
.;:: 550 750
§ 500 c>~'O'/o-------~ 700
~ 450
6SO
4OO0'--~2~4-""'6~8-1""'0---'12 600
0 2
Number of Distractors Number of Distractors
Figure S. Left, middle: Reaction time to judge the direction of the vernier offset of the odd-spatial-frequency
target plotted apinst the number of dimactors in the display. Note the different values on the y-axes of these graplLoi.
Open circles represeut data from the comistent-mapping experiment; filled circles represent data from the variable-
mapping experiment. Error bars show one standard error. Right: Average error rates for the three observers.
interchanged unpredictably from trial to trial. And again,
this effect cannot be ascribed to a speed-accuracy tradeoff
since there is no significant correlation between error rate
and reaction time.
This second pair of experiments employed very differ-
ent stimuli than did the previous experiments, and yet the
same overall pattern of results was obtained. When the
unique feature of the target was consistent across trials,
reaction times to identify a subtle feature of the target were
unaffected by the number of distractors. When the unique
feature of the target reversed unpredictably with the dis-
tractor feature, reaction times decreased as the number
of distractors increased.
GENERAL DISCUSSION
The pattern of our reaction time results reveals fun-
damental differences in the processes invoked by differ-
ent visual-search tasks. One difference can be seen in the
results of the three experiments involving different judg-
ments under the variable-mapping condition. Reaction
times for the discrimination of a subtle feature of the tar-
get decrease as the number of distractors increases, while
reaction times to detect the target or identify its unique
feature are unaffected by the number of distractors. This
discrepancy, we believe, is due to the involvement of fo-
cal attention in the first case and distributed attention in
the others. A second difference among the processes in-
volved in different visual-search tasks can be seen in the
results of the two experiments involving shape identifi-
cation under different mapping conditions. If we assume
that the shape judgment requires focal attention, then these
results suggest the existence of two classes of mechanisms
by which focal attention can be directed to the target.
These are (l) a largely top-down mechanism that requires
knowledge of the target and (2) a largely bottom-up mech-
anism that can operate without such knowledge. As dis-
cussed earlier, current models of bottom-up mechanisms
predict our finding that reaction times falloff sharply as
the number of distractors increases. The models do not
predict the results of the detection- and unique-feature-
0.125
RO
0.100
...... - ..... _--
Q /0......... - ..
0.G75
\
\
~"CJ'
I
I .......
.......
'0
0.050
0.025
4 6 8 10 12
O.OOO'--~~~~-~~
o 2 4 6 8 10 12
Number of Distraetors
identification tasks, and in the following paragraphs, we
consider the processes that might be involved in these tasks.
We consider first the task of determining the color of
the odd-colored target. There are several strategies that
an observer could use to judge the color of the target. One
strategy is to determine which color has the smaller num-
ber of occurrences in the display. Increasing the number
of distractors in the display would increase the difference
in the number of occurrences of the two colors and
presumably facilitate this decision. A similar strategy is
to determine which color has more than one occurrence
in the display and then to choose the other color. Again,
it is reasonable to expect that increasing the number of
distractors in the display would facilitate this decision.
However, we found that the reaction times to judge the
color of the target were unaffected by the number or
distractors.
It is also not clear how observers detected the presence
of an odd target in the detection experiments. Processes
similar to the attention-guiding mechanisms outlined in
the introduction have been suggested. In the consistent-
mapping/detection experiment, the observer could respond
on the basis of the presence or absence of activity in the
appropriate feature map (freisman & Gelade, 1980). And
in the variable-mapping/detection experiment, observers
could use one of the bottom-up models to detect the
presence of an odd element (Julesz, 1986; Koch & Ullman,
1985). But if the computations involved in these bottom-
up mechanisms are local, then search times should be sen-
sitive to the number of distractors in the display. Although
there are numerous reports that the number of distractors
has no effect on searching for a readily segregated tar-
get, Bacon and Egeth (1991) recently reported that de-
tection times may decrease as the number of distractors
increases. In our experiments, however, the same condi-
tions that produced a large effect for the shape judgment
produced no effect for the detection judgment. Thus, our
observers are not using the same processes to find the tar-
get for detection and shape identification. It also appears
that the process they use for detection is insensitive to the
number of distractors.
 ROLE OF ATTENTION IN VISUAL SEARCH 471

We favor an alternative explanation of the detection and
unique-feature-identification results that relates to coarse-
scale pattern matching. We assume that each of our tasks
involves the postattentive processes of pattern recogni-
tion but that the particular pattern that is to be recognized
depends on the observer's judgment. Sometimes the pat-
tern is that of an individual element, as in the shape judg-
ment; other times it is that of the whole display, as in the
detection and unique-feature-identificationjudgment. Be-
cause it is the attentional window that selects the infor-
mation for pattern recognition, these different tasks re-
quire attention to operate in different ways. It is these
different attentional operations that are reflected in the
reaction time results. When the display as a whole is used
for pattern recognition, reaction time is unaffected by the
number of distractors in the display. When only a part
of the display is used, then attention must be directed to
the appropriate part.
There are several ways attention may be directed, and
each is associated with a characteristic pattern of reac-
tion time results. We have already discussed top-down
mechanisms, which guide attention to a particular feature
(flat-search function), and bottom-up mechanisms, which
guide attention to an odd target (negative-search function).
There is a third type of attentional guidance that we did
not see in these experiments but that has been reported
by many others. A serial search through all elements in
the display is required if discriminating between the tar-
get and distractor requires focal attention. This type of
search produces positive search functions (Estes & Wessel,
1966; Treisman & Gelade, 1980).
In closing, we refer the reader to a general framework
of vision proposed by Nakayama (1990) that has greatly
influenced our thinking about these experiments. The
model has three components, as depicted in Figure 6. On
the left of this figure, incoming sensory information is
encoded by a massively parallel system that represents
multiple features at multiple scales. This system has a
pyramid-like structure, with each level in the pyramid re~
resenting the visual field at a different spatial scale. The
base of the pyramid is a very fine scale representation,
and the apex is a very coarse representation. The amount
of information contained at any level is directly related
to the resolution of that level.
On the right in Figure 6 is visual memory, an associa-
tive network of icons or templates. Pattern recognition
is accomplished by activating an icon in visual memory.
It is assumed that individual icons have a low informa-
tion content. This assumption is based in part on the results
of reading studies that assess the amount of information
that is processed simultaneously (Legge, Pelli, Rubin, &
Schleske, 1985; Rayner, 1978).
In the middle in Figure 6 is selective attention, the link
between the massive amount of information in the pyramid
and the small amount of information in an icon. Atten-
tion can be thought of as an information bottleneck through
which only a small set amount of information may pass.
Attention-guiding mechanisms are required to select the
appropriate information from the pyramid to be passed
on to visual memory.
This model offers an explanation of how the tradeoff
between spatial extent and resolution is accomplished prior

Pyramid Attention Visual Memory

"·••
.--
o 0
00
attention is guided by
top-down and bottom-up
mechanisms
o
o
o

Figure 6. A schematic of the model proposed by Nakayama (1990) applied to the present results. Sensory
infonnation is encoded in a multiresolution pyramid. Attention selects the infonnation from the pyramid
that is to be used for pattern matching. The continUOWlIine depicts the selection of infonnatioD for the shape-
discrimination judgment. Attention-guiding mechanisms are required to direct attention to the appropriate
location within a high-resolution level of the pyramid. The hatched line indicates the selection of infonna-
tion for the detection judgment.
472 BRAVO AND NAKAYAMA
to pattern recognition. Depending on the observer's task,
detailed information about a small spatial region may be
selected from a lev"l close t(1. the base of the pyramid or,
alternatively, information about the coarsest scene fea-
tures may be selected from a level close to the apex. In
our experiments, observers knew before every trial the
judgment, and thus the resolution, that would be required.
This allowed them to select the appropriate level within
the pyramid before each trial. However, they did not know
the location of the target within the stimulus, and thus
when high resolution was required, they had to draw on
attention-guiding mechanisms to select the appropriate lo-
cation within the high-resolution level of the pyramid. It
is this latter step that is reflected in the unusual reaction
time results reported here.
REFERENCES
ALLMAN, J., MIEZEN, F., Ill: McGUINESS, E. (1985). Stimulus specific
responses from beyond the classical receptive field: Neurophysiological
mechanisms for local-global comparisons in visual neurons. Annual
Review of Neuroscience, 8, 407-430.
BACON, W., Ill: EGETH, H. (1991). Local processes in preattentive fea-
ture detection. Journal ofExperimental Psychology: Human Percep-
tion & Performance, 17,77-90.
BERGEN, J., Ill: JULESZ, B. (1983). Rapid discrimination of visual pat-
terns. I. E. E.E. Transactions on Systems, Man & Cybernetics, SMC-13,
857-863.
DoNDERI, D. C., Ill: ZELNICKER, D. (1969). Parallel processing in visual
same-different decisions. Perception & Psychophysics, 5, 197-200.
DUNCAN, 1., Ill: HUMPHREYS, G. W. (1989). Visual search and stimu-
lus similarity. Psychological Review, 96, 433-458.
EGETH, H., JONIDES, J., Ill: WALL, S. (1972). Parallel processing of multi-
element displays. Cognitive Psychology, 3, 674-698.
ERIKSEN, C. W., Ill: ST. JAMES, J. D. (1986). Visual attention within
and around the field of focal attention: A zoom lens model. Percep-
tion & Psychophysics, 40, 225-240.
EsTES, W. K., '" WESSEL, D. L. (1966). Reaction time in relation to
display size and correctness of response in forced-ehoice visual sig-
nal detection. Perception & Psychophysics, I, 369-373.
FISCHER, B., '" BREITMEYER, B. (1987). Mechanisms of visual atten-
tion revealed by saccadic eye movements. Neuropsychologia, 25,
73-83.
GRONER, R., '" GRONER, M. T. (1989). Attention and eye movement
control: An overview. European Archives ofPsychiatry & Neurological
Sciences, 239, 9-16.
JULESZ, B. (1986). Texton gradients: The texton theory revisited. Bio-
logical Cybernetics, 54, 245-251.
KOCH, C., '" ULLMAN, S. (1985). Shifts in selective visual attention:
Towards the underlying neural circuitry. Human Neurobiology, 4,
219-227.
LABERGE, D., Ill: BROWN, V. (1989). Theory of attentional operations
in shape identification. Psychological Review, 96, 101-124.
LEGGE, G., PELLI, D., RUBIN, G. S., '" ScHLESKE, M. M. (1985).
Psychophysics of reading. I. Normal vision. Vision Research, 25,
239-252.
MAYFRANK, L., MOBASHERY, M., KIMMlG, H., '" FISCHER, B. (1986).
The role of fixation and visual attention in the occurrence of express
saccades in man. European Archives of Psychiatry & Neurological
Sciences, 235, 269-275.
NAKAYAMA, K. (1990). The iconic bottleneck and the tenuous link be-
tween early visual processing and perception. In C. Blakemore (Ed.),
Vision: Coding and efficiency (pp. 411-422). Cambridge: Cambridge
University Press.
NEISSER, U. (1963). Decision-time without reaction time: Experiments
in visual scanning. American Journal of Psychology, 76, 376-385.
POSNER, M. (1980). Orienting of attention. Quanerly Journal of Ex-
perimental Psychology, 32, 3-25.
RAYNER, K. (1978). Eye movements in reading and information pro-
cessing. Psychology Bulletin, 85, 618-660.
REMINGTON, R. W. (1980). Attention and saccadic eye movements. Jour-
nal ofExperimental Psychology: Human Perception & Performance,
6,726-744.
ScHNEIDER, W., '" SHIFFRIN, R. M. (1977). Controlled and automatic
human information processing. I. Detection, search, and attention.
Psychological Review, 84, 1-66.
TREISMAN, A. (1988). Feature and objects: The fourteenth Bartlett me-
morial lecture. Quanerly Journal ofExperimental Psychology, 4OA,
201-237.
TREISMAN, A., Ill: GELADE, G. (1980). A feature-integration theory of
attention. Cognitive Psychology, 12, 97-136.
WOLFE, J. M., CAVE, K. R., Ill: FRANZEL, S. L. (1989). Guided search:
An alternative to the feature integration model of visual search. Journal
of Experimental Psychology: Human Perception & Performance, 15,
419-433.
YANTIS, S., Ill: JONIDES, J. (1984). Abrupt visual onsets and selective
visual attention: Evidence from visual search. Journal ofexperimen-
tal Psychology: Human Perception & Performance, 10, 601-621.
NOTES
I. By practiced observers, we mean observers who have participated
in a variety of experiments and are familiar with psychophysical proce-
dures. We have not included the observations from M.B. (the first author)
because in the process of developing the test described in the present
research, she had large and differing amounts of practice on each task.
In addition, in using her as an observer over many trials, we found that
some of the differences in reaction time described in this paper decrease
with very prolonged practice.
2. Gabor patches are spatially localized sine-wave gratings, whose
modulation envelope is a two-dimensional Gaussian function. The Gauss-
ian window of the Gabor patches had a width of 0.62° when the enve-
lope's amplitude was lie, and the suprathreshold area of the Gabor
patches subtended approximately 2°.
(Manuscript received August 19, 1991;
revision accepted for publication December 27, 1991.)